Scholarlink

<p>In today’s digital era, social media platforms have become an integral part of everyday life, enabling the free exchange of ideas and opinions. However, the widespread presence of offensive content on these platforms can lead to significant psychological and social harm. While previous studies have predominantly focused on Twitter data and conventional deep learning models for offensive language detection, this study addresses the problem using real-world Instagram data. A novel hybrid approach is proposed, combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, with hyperparameter optimization guided by the Grey Wolf Optimizer (GWO). This integration of deep learning techniques with a nature-inspired metaheuristic optimization algorithm enables automatic hyperparameter tuning, helps avoid local optima, and enhances generalization performance. For feature extraction, word embedding techniques such as Glove and Word2Vec were utilized. The resulting feature vectors were processed through CNN layers, stacked LSTM layers, and fully connected layers to classify content into three categories: offensive, hateful, and neutral. Experimental results demonstrated that the proposed GWO-CNN-LSTM hybrid model outperformed baseline models including standalone LSTM, CNN, and GWO-LSTM in terms of accuracy (88.60%), precision (88.59%), recall (88.60%), and F1-score (88.49%). The use of authentic Instagram data further contributes to the robustness and adaptability of the model in diverse and dynamic linguistic contexts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detecting offensive language on instagram with a combined approach of the Gray Wolf algorithm and deep learning networks

  • 期刊论文

  • 发表日期：

    2025年9月25日

查看全文
  • Homa Omarzadeh,
  • Monireh Hosseini

摘要

In today’s digital era, social media platforms have become an integral part of everyday life, enabling the free exchange of ideas and opinions. However, the widespread presence of offensive content on these platforms can lead to significant psychological and social harm. While previous studies have predominantly focused on Twitter data and conventional deep learning models for offensive language detection, this study addresses the problem using real-world Instagram data. A novel hybrid approach is proposed, combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, with hyperparameter optimization guided by the Grey Wolf Optimizer (GWO). This integration of deep learning techniques with a nature-inspired metaheuristic optimization algorithm enables automatic hyperparameter tuning, helps avoid local optima, and enhances generalization performance. For feature extraction, word embedding techniques such as Glove and Word2Vec were utilized. The resulting feature vectors were processed through CNN layers, stacked LSTM layers, and fully connected layers to classify content into three categories: offensive, hateful, and neutral. Experimental results demonstrated that the proposed GWO-CNN-LSTM hybrid model outperformed baseline models including standalone LSTM, CNN, and GWO-LSTM in terms of accuracy (88.60%), precision (88.59%), recall (88.60%), and F1-score (88.49%). The use of authentic Instagram data further contributes to the robustness and adaptability of the model in diverse and dynamic linguistic contexts.